Multi‐Night Electroencephalography Reveals Positive Association Between Sleep Efficiency and Hippocampal Subfield and Entorhinal Cortex Volumes in Healthy Aging Authors: Attila Keresztes, Éva M Bankó, Noémi Báthori, Vivien Tomacsek, Virág Anna Varga, Ádám Nárai, Zsuzsanna Nemecz, Ádám Dénes, Viktor Gál, Petra Hermann, Péter Simor, Zoltán Vidnyánszky Published: 2024 ABSTRACT ABSTRACTAge‐related atrophy of the human hippocampus and the enthorinal cortex starts accelerating at around age 60. Due to the contributions of these regions to many cognitive functions seamlessly used in everyday life, this can heavily impact the lives of elderly people. The hippocampus is not a unitary structure, and mechanisms of its age‐related decline appear to differentially affect its subfields. Human and animal studies have suggested that altered sleep is associated with hippocampal atrophy. Yet, we know little about subfield specific effects of altered sleep in healthy aging and their effect on cognition. Here, in a sample of 118 older middle‐aged and older adults (M age = 63.25 y, range: 50–80 y), we examined the association between highly reliable hippocampal subfield and entorhinal cortex volumetry (n = 112), sleep measures derived from multi‐night recordings of portable electroencephalography (n = 61) and episodic memory (n = 117). Objective sleep efficiency—but not self‐report measures of sleep—was associated with entorhinal cortex volume when controlling for age. Age‐related differences in subfield volumes were associated with objective sleep efficiency, but not with self‐report measures of sleep. Moreover, participants characterized by a common multivariate pattern of subfield volumes that contributed to positive sleep–subfield volume associations, showed lower rates of forgetting. Our results showcase the benefit of objective sleep measures in identifying potential contributors of age‐related differences in brain‐behavior couplings.Keywords: Dreem, episodic memory, hippocampal subfields, partial least squares correlation, sleep duration Summary Six nights of portable EEG data collected at participants' home. Six nights of portable EEG data collected at participants' home. Entorhinal cortex volume was specifically associated with objective sleep efficiency. Entorhinal cortex volume was specifically associated with objective sleep efficiency. Objective but not subjective sleep efficiency was associated with age differences in hippocampal subfield and entorhinal cortex volumes. Objective but not subjective sleep efficiency was associated with age differences in hippocampal subfield and entorhinal cortex volumes. Positive sleep–hippocampal subfield volume coupling was associated with lower forgetting. Positive sleep–hippocampal subfield volume coupling was associated with lower forgetting. Introduction Volume loss of the hippocampus and surrounding mediotemporal cortical areas during human aging is accelerated relative to most other cortical areas (Fjell et al. 2009; Pomponio et al. 2020; Raz et al. 2010). In vivo animal and post‐mortem human studies have suggested that these volumetric changes can reflect a combination of factors (for a comprehensive review, see Bettio, Rajendran, and Gil‐Mohapel 2017) including decreased number, size, and density of axons and dendrites due to synaptic pruning, to some extent decreased number of neurons (Lister and Barnes 2009; West 1993), as well as changes in astrocytic and glial cell size and density (Ojo et al. 2015). These alterations, in turn, are thought to be the consequences of genetically (Guo et al. 2019) and environmentally (Binnewies et al. 2023) driven changes in blood supply (Shing et al. 2011), in hormonal (McEwen 1997), neurotrophic (Buhusi et al. 2017), and neuron–glia interaction regulation (Ojo et al. 2015), and in neurogenesis (Boldrini et al. 2018; Sorrells et al. 2018; Spalding et al. 2013). Understanding the mechanisms of these changes is crucial for mitigating or potentially reversing (Bettio, Rajendran, and Gil‐Mohapel 2017; Kim et al. 2016; McEwen 1997) their adverse effects on cognitive functions supported by the hippocampus. Because these functions—including memory (Moscovitch et al. 2016; Wixted and Squire 2011), navigation (Eichenbaum 2017; Whittington et al. 2020), and even language (Duff et al. 2020)—have major roles in everyday life, their impairment has a significant impact in healthy, and devastating effect in pathological aging. Critically, hippocampal volumetry is currently the only widely accessible, valid, and reliable method to assess hippocampal atrophy in humans in vivo. Importantly, age‐related changes driving volumetric loss of the hippocampus appear to be expressed differentially across its subfields (Bartsch and Wulff 2015) varying in cytoarchitecture, macrostructure, connectivity (Amaral and Lavenex 2007), and function (Genon et al. 2021; Mueller et al. 2011). This heterogeneity may explain the fact that although hippocampal volume loss has been consistently detected in aging, no consistent trajectories of hippocampal subfield volume loss have been identified (de Flores, La Joie, and Chételat 2015). Thus, for a mechanistic understanding of hippocampal atrophy in healthy aging, it is crucial to tease apart subfield specific contributions of factors affecting hippocampal volume. One such relatively understudied factor is sleep. It is well documented that sleep patterns altered in apnea (Kim et al. 2016; Macey et al. 2002) and other sleep breathing disorders (Marchi et al. 2020), insomnia (Noh et al. 2012), narcolepsy (Joo et al. 2012), as well as in pathological conditions such as depression (Campbell et al. 2004; Videbech and Ravnkilde 2004) are associated with smaller total hippocampal volume. Recent studies have also provided evidence suggesting that sleep disturbances may partly underlie the effects of pathological (Burke et al. 2022; Elcombe et al. 2015) and healthy (Alperin et al. 2019; Carvalho et al. 2017; Fjell et al. 2020; Liu et al. 2018) aging on hippocampal volume with subfield specific effects (André et al. 2023; De Looze et al. 2022; Joo et al. 2014; Lam et al. 2021; Liu et al. 2021; Macey et al. 2018). However, the absence of a consistent pattern of observed sleep–volume associations in healthy aging when only total hippocampal volumes are assessed underscores the need for subfield specific studies (De Looze et al. 2022; Sabeti et al. 2018). A few recent studies have assessed hippocampal subfield volumes as a function of sleep in healthy aging (André et al. 2023; De Looze et al. 2022; Liu et al. 2021; Martineau‐Dussault et al. 2022). In a large sample (n = 417; M age = 68.8 y, SD age = 7.4 y), De Looze et al. (2022) found that both too short and too long sleep duration were associated with smaller subiculum volume, with too long sleep additionally associated with smaller volumes of the cornu ammonis 1 (CA1) subfield. Somewhat consistently, Liu et al. (2021) in a smaller sample study (n = 67; M age = 70 y, SD age = 5.8 y) found that healthy participants with poor sleep had smaller volumes of subiculum, CA1, and dentate gyrus. However, in a cognitively unimpaired subsample of participants with and without obstructive sleep apnea (n = 68; M age = 68 y, SD age = 7.2 y), Martineau‐Dussault et al. (2022) found no association between hippocampal subfields and sleep hypoxemia. Importantly, these studies either relied on self‐report measures to assess sleep quality and duration (De Looze et al. 2022; Liu et al. 2021) and/or assessed volumes relying on low (1 mm) in‐plane resolution scans (De Looze et al. 2022; Martineau‐Dussault et al. 2022) with limited validity for hippocampal subfield volumetry (Wisse, Biessels, and Geerlings 2014; Wisse et al. 2021). Thus, it remains to be clarified to what extent the observed associations with hippocampal subfield volumetry reflect alterations of objective sleep parameters. The only study to date that used submillimeter resolution structural MRI together with polysomnography to study subfield volume–sleep association in a sample (n = 122; M age = 69 y, SD age = 3.9 y) of cognitively unimpaired older adults (André et al. 2023) found that entorhinal cortex, subiculum, CA1 and dentate gyrus volumes negatively correlated with sleep apnea severity only in a subsample (n = 26) of amyloid positive participants. André and colleagues additionally tested for and detected no association between episodic memory, assessed by the California Verbal Learning Test, and sleep, but did find a negative association between CA1 volume and memory. Importantly, this study used at home EEG recordings of a single night lowering reliability of its results due to first‐night effects of sleep measurement (Browman and Cartwright 1980). To address these gaps, we examined the association between highly reliable semi‐automatic hippocampal subfield volumetry and sleep measures derived from multi‐night recordings of portable electroencephalography (EEG) and their association with standard measures of learning and memory in a sample of community dwelling older middle‐aged and older adults. We hypothesized that (1) age is negatively associated with subfield volumes, sleep uality and memory, (2) sleep quality is positively associated with hippocampal subfield and entorhinal cortex volumes, (3) age differences in subfield volumes are negatively associated with sleep quality, and (4) individual differences in sleep quality–volume associations manifest in memory performance. Finally, because prior studies have suggested sex as a biological variable affecting sleep (Mong and Cusmano 2016; Won et al. 2020), volumes of mediotemporal regions and memory (Asperholm et al. 2019; Jack Jr et al. 2015; Malykhin et al. 2017; Mosconi et al. 2021) across the adult lifespan, we additionally explored sex effects in all analyses. Materials and Methods Participants Community dwelling older middle‐aged and older adults were recruited as part of the Hungarian Longitudinal Study of Healthy Brain Aging (HuBA) at the Brain Imaging Centre of the Research Centre for Natural Sciences in Budapest, Hungary (Bankó et al. 2024). They had no previous or current diagnosis of neurological, psychiatric or sleep disorders, untreated hypertension, diabetes, and no history of alcohol or drug abuse or malignant tumor in the past 5 years and were screened using the Mini mental state examination (MMSE) for dementia (cut‐off: 26). Sleep disorders were screened by one general question on the presence of diagnosed sleep disorders, and a conversation on potential sleep complaints during one‐on‐one training sessions for the subsample undergoing at‐home sleep EEG. Data of one hundred and eighteen 50–80‐year‐old participants (M age = 63.25 y, SD age = 7.23 y, 68 females, M education = 16.25, SD education = 2.55), who had hippocampal subfield data (n = 112) or sleep recordings for at least 3 nights (n = 61), were included in this study (Figure 1). Of these, 55 participants (M age = 62.64, SD age = 6.85, 30 females) had both hippocampal subfield and sleep data. The study was approved by the National Institute of Pharmacy and Nutrition (OGYÉI/68903/2020), and all participants gave written informed consent. Demographic data of included participants together with covariates are reported in Table 1. (A) Flowchart of the study. One hundred twenty‐six participants were invited to the MRI session, of these 121 were offered the option to participate in the sleep assessment using at‐home sleep EEG. Sixty‐nine opted in. (B) Overlap of available data. The intersection shows that 55 participants had all data types. TABLE 1Descriptives of demographic variables and covariates.Full sample (58% [n = 68] female)Sleep EEG sample (52% [n = 32] female) n M SD n M SD Age (years)11863.257.236162.206.89Education (years)10916.252.556116.112.27IQ (score)103121.7512.1260124.4310.94MET (minutes)116347.25353.4460350.8403.97Systolic BP (mmHg)98133.5914.5454133.0414.93Diastolic BP (mmHg)9882.818.285483.878.80Depression (score)1152.842.89592.752.82SUB (mm3)112789.77129.8955793.82130.5CA1‐2 (mm3)112449.9462.7155448.2267.27DG‐CA3 (mm3)112574.2894.4255582.1299.09EC (mm3)112519.2367.3155524.2770.76OSE (%)6188.654.356188.654.35SSE (%)11086.2213.175987.4912.9GSQS (score)6718.238.116118.228.57AIS (score)804.193.12594.222.95RAVLT recall (items)1179.773.39619.793.2RAVLT forgetting (items)1172.442.29612.462.26Abbreviations: AIS: Athens Insomnia Scale, BP: blood pressure, CA1‐2: cornu ammoni regions 1–2, DG‐CA3: dentate gyrus‐CA3, EC: entorhinal cortex, GSQS: Groningen Sleep Quality Scale, MET total metabolic equivalent (MET) minutes per week, OSE: objective sleep efficiency, Recall: number of recalled items on the Rey Auditory Verbal Learning Test (RAVLT)'s delayed recall test, Forgetting: number items learnt but not recalled during the delayed recall test of the RAVLT, SSE: subjective sleep efficiency, SUB: subiculum. Descriptives of demographic variables and covariates. Abbreviations: AIS: Athens Insomnia Scale, BP: blood pressure, CA1‐2: cornu ammoni regions 1–2, DG‐CA3: dentate gyrus‐CA3, EC: entorhinal cortex, GSQS: Groningen Sleep Quality Scale, MET total metabolic equivalent (MET) minutes per week, OSE: objective sleep efficiency, Recall: number of recalled items on the Rey Auditory Verbal Learning Test (RAVLT)'s delayed recall test, Forgetting: number items learnt but not recalled during the delayed recall test of the RAVLT, SSE: subjective sleep efficiency, SUB: subiculum. Hippocampal Subfield and Entorhinal Cortex Volumetry MRI Acquisition We acquired high‐resolution proton density (PD) weighted turbo spin echo images (TR = 6490 ms; TE = 16 ms; FOV =206 × 16 × 66 mm; resolution = 0.4 × 0.4 × 2 mm; 32 slices, acquisition time = 4 min 15 s) of the hippocampus and surrounding mediotemporal areas oblique to the coronal plane, perpendicular to the longitudinal axis of the right hippocampus, as well as high‐resolution 3D T1‐weighted images using a Magnetization Prepared RApid Gradient Echo (MPRAGE) sequence (TR =2300 ms; TE = 3.03 ms; FOV = 256 × 256 × 192 mm; resolution = 1 mm isotropic; acquisition time = 5 min 21 s) on a 3 T Siemens Magnetom Prisma scanner. Hippocampal Subfield and Entorhinal Cortex Segmentation Following (Bender et al. 2018), in a semi‐automatic procedure, we used the automated segmentation of hippocampal subfields (ASHS) (Yushkevich et al. 2015) software package to delineate four regions of interest (ROIs) bilaterally based on a custom lifespan atlas (Bender et al. 2018). ASHS uses a multi‐atlas method that integrates information from manual segmentations and thus provides higher precision—and inherent validity—at the individual level than other approaches using a single‐atlas (e.g., Freesurfer). The manual delineation procedure used to create the custom atlas has been described in detail in (Bender et al. 2018; Keresztes et al. 2017). In brief, we delineated four ROIs: three within the hippocampal body—subiculum (SUB), CA regions 1 and 2 (CA1‐2), and dentate gyrus–CA3 (DG‐CA3)—as well as the entorhinal cortex (EC) on six consecutive slices anterior to the hippocampal body (see Figure 2A). We constrained analyses of SUB, CA1‐2, and DG‐CA3 to the hippocampal body because of the questionable validity of currently available methods to delineate these subfields in the head or the tail of the hippocampus (Wisse et al. 2017). Similar validity concerns motivated our decision to use an atlas that collapses smaller subfields into larger ROIs: DG and CA3 into DG‐CA3, CA1, and CA2 into CA1‐2, and presubiculum, subiculum, and parasubiculum into SUB; reliable and valid delineation of these separate subfields within the above ROIs on high‐resolution images acquired at 3 T are yet to be established (Keresztes et al. 2020; Wisse et al. 2017). Hippocampal body slices were defined as slices ranging from the most anterior slice on which the uncus of the hippocampus was no more clearly visible to the most posterior slice where the lamina quadrigemina (i.e., the inferior and superior colliculi; circled) were still clearly visible (see Keresztes et al. 2020). Hippocampal body ranges were manually defined by a rater (V.A.V.). This rater had been first trained on an independent set of PD‐weighted images (7 sets of 12 brains) acquired with identical scanning parameters to the PD‐weighted images used in this study until she reached good inter‐rater reliability (all Cohen's kappas > 7) with a rater with > 8 years of experience in manual segmentation of hippocampal subfields (A.K.). Then she ranged all brains. Finally, V.A.V. reranged 12 brains after a 2‐month interval to establish intra‐rater reliability (all Cohen's kappas > 0.77). Segmentations were visually inspected by both raters who then corrected obvious errors. Volumes for all regions were adjusted for intracranial volume, estimated by ASHS, based on the analysis of covariance approach (Jack et al. 1989; Raz et al. 2005), then summed across hemispheres for all analyses. (A) Example hippocampal subfield segmentation of the right hippocampus and entorhinal cortex into four subfields: Subiculum (orange), dentate gyrus—Cornu Ammoni 3 (DG‐CA3; green), CA1‐2 (blue), and entorhinal cortex (red). (B) Negative age‐associations of the four subfields, non significant only for DG‐CA3. Sleep Data EEG Data Acquisition Sleep recordings were acquired at participants' homes using a wireless Dreem2 EEG headband (Dreem, Rythm, Paris, France; DH) with five electrode sensors yielding seven bipolar derivations at a 250 Hz sampling frequency. Participants wore the DH for at least 7 consecutive nights starting an average of 29.9 days (2–193, SD = 33.6) after the MRI session. Participants received instructions on the sleep study protocol, covering headband and proper electrode placement, cleaning techniques, and smartphone app use for data uploads during one‐on‐one meetings prior to starting the seven‐day sleep study phase. They received isopropyl alcohol wipes and a sweatband to enhance signal quality by maintaining electrode contact as well as a detailed user manual. In addition, they had access to educational videos and 24/7 technical support via email, phone, and online meetings. If necessary, additional meetings were scheduled after a prompt quality check on first night data. EEG Data Preprocessing Sleep recordings were qualified and scored by an automated deep learning based sleep stage analysis based on American Academy of Sleep Medicine standards with high accuracy (Arnal et al. 2020), akin to a qualified PSG scorer. This automatic analysis ensures agreement between sleep staging decisions across epochs and recordings, whereas manual sleep stage scoring can vary among sleep experts, leading to greater inter‐rater variability (Arnal et al. 2020; Danker‐Hopfe et al. 2009; Lee et al. 2022). However, noise and artifacts may affect results. The algorithm assigns a quality index to each EEG channel, representing the signal‐to‐noise ratio. A qualified PSG expert (P.S.) visually assessed 20 randomly selected whole‐night (> 5‐h) recordings using visual EEG scoring rules (Berry et al. 2015) to establish the hypnograms and compare them to those provided by the algorithm to identify a quality threshold for our study. The agreement between automatic and manual scorings ranged between 0.79 and 0.90 (wake: 0.64–0.85, Stage 1: 0.11–0.77, Stage 2: 0.81–0.96, slow wave sleep: 0.59–0.94, REM: 0.67–0.98) if signal quality was at least 75% in at least one channel but dropped below 0.60 if signal quality was poorer. Thus, we decided to constrain further analyses to recordings in which signal quality exceeded a threshold of 75% on at least one channel. Participants recorded 460 nights altogether, from which we included 365 nights in our analysis, resulting in a 21% dropout rate. As a final step, we ran two analyses to test for first night effects within the included nights: A paired samples t‐test between objective sleep efficiency of the first and second nights (t(46) = −1.66, p = 0.105), and a linear mixed model assessing the effect of night order on sleep efficiency (β = 0.25, SE = 0.16, t(289.34) = 1.55, p = 0.123). Based on these, we did not discard any further nights. Of the nights included, 5 participants had three, 8 had four, 6 had five, 11 had six, 26 had seven, and 5 had eight nights. This procedure yielded an average of 5.98 nights (SD = 1.47, Mdn = 7), with a quality index of 93.04% (SD = 8.47, Mdn = 95.37). We used the proportion of the time spent asleep and the time spent in bed, that is from lights out to lights up (objective sleep efficiency) as an independent variable. (Due to the large variation in validity across sleep stages, sleep architecture data and its associations with our dependent variables are presented in exploratory analyses only in the [Table S4]). Of note, we did not differentiate between nights before workdays and nights before non‐workdays. Subjective Measures of Sleep To assess subjective estimates of sleep efficiency and quality, we used three measures derived from self‐report questionnaires. First, we calculated subjective sleep efficiency similarly to the objective sleep efficiency using items 1, 3, and 4 of the Pittsburgh Sleep Quality Index (PSQI; Buysse et al. 1989; Hungarian adaptation: Takács et al. 2016). Second, we assessed subjective sleep quality for the past month using the Athens Insomnia Scale (Hungarian adaptation: Novák 2004; AIS; Soldatos, Dikeos, and Paparrigopoulos 2000). The 8‐item AIS assesses the severity of sleep problems based on the insomnia criteria defined by the International Classification of Diseases (ICSD‐10; ICSD: American Academy of Sleep Medicine 2005) with a total score of 6 or above indicating the presence of insomnia symptoms. We used the 14‐item Groningen Sleep Quality Scale (GSQS; Meijman et al. 1988; Hungarian adaptation: Simor et al. 2009) to assess the subjective quality of sleep each morning, after awakening. Higher scores in the GSQS reflect higher subjective sleep fragmentation. We modified the original binary (yes/no) scoring of the GSQS to a 5‐point Likert scale to weight each statement and used an average of GSQS scores across all nights. Memory Measures We used the Rey Auditory Verbal Learning Test (RAVLT; Lezak 2004; Hungarian adaptation: Kónya and Verseghi 1995) to assess delayed episodic memory and forgetting. In the learning phase of the RAVLT participants are presented auditorily with a list of 15 nouns (trial duration: 1 s per item) and are then instructed to freely recall all words they remember. The learning phase consists of 5 such encoding‐retrieval blocks. This is followed by one identical encoding‐retrieval block with a novel list of 15 words to assess interference effects (data not used here). Finally, after a 30‐min delay period a final free recall is administered. As measures of episodic memory, we used the number of items recalled in delayed recall as well as the number of items forgotten relative to the last free recall of the learning phase. Covariates Beyond age (in days), sex, and education (years spent in education), we assessed several covariates with known associations with cognition and/or hippocampal and entorhinal cortical volumes: Physical activity total metabolic equivalent minutes (MET) per week measured by the short version of the International Physical Activity Questionnaire (Craig et al. 2003), total IQ (measured by the Wechsler Adult Intelligence Scale 4th Edition (WAIS‐IV; Pearson PLC, London, UK)), depression measured by the short version of the Geriatric Depression Scale (GDS; Sheikh and Yesavage 1986), and blood pressure (systolic and diastolic blood pressure measured after each scanning session). Descriptives of covariates and their association with age are reported together with demographic variables in Table 1 and Figure S1, respectively. Statistical Analyses We conducted five sets of analyses. First, to explore associations of age with our main variables of interest—four volume measures (SUB, CA1‐2, DG‐CA3, and EC), four sleep variables (objective sleep efficiency, subjective sleep efficiency, AIS, and GSQS), and two memory variables (delayed recall and forgetting), we ran 10 bivariate zero‐order Pearson's correlations with significance level corrected for multiple comparison being α = 0.005. Second, we tested whether sleep measures predicted subfield volumes in regression models defined separately for each sleep measure as independent variables and each subfield volume as dependent variable. To this end, we ran 16 (4 sleep measures × 4 subfield measures) regression models including all covariates listed in Section 2.5. To test model fit of these regressions, significance level was corrected for multiple comparisons with α = 0.05/16 = 0.0031. Within significant models, the level of alpha to assess significance of individual predictors was corrected for multiple comparisons with α = 0.05/9 = 0.0055. (Due to validity concerns described above (Section 2.3.2), sleep architecture data and its associations with subfield volumes are presented in exploratory analyses only in the [Table S4]). Third, to test for the association between age‐related differences in hippocampal subfield and entorhinal cortex volumes and sleep, we conducted a partial least squares analyses (PLSC; Krishnan et al. 2011). Partial least squares correlations (PLSC) is a robust method that allows detecting multivariate associations between groups of variables with even high collinearity (McIntosh and Lobaugh 2004). PLSC has been successfully used in cross‐sectional lifespan developmental studies to extract a latent profile of brain variables that share maximum variance with age (Keresztes et al. 2017; Muehlroth et al. 2020). Such a latent variable allows for increasing the precision of cross‐sectional studies that estimate a process of change via sampling from participants of different ages. Here, we used PLSC to assess a latent decline of subfields via calculating a latent variable that maximally expressed the shared variance between age and volumes on the subsample of participants (n = 112) having hippocampal subfield data. Briefly, we decomposed the correlation matrix a vector containing age and a 4‐column matrix of subfield volumes using singular value decomposition. This produces a left singular vector of age weights (U), a right singular vector of subfield weights (V), and a diagonal matrix of singular values (S). From this decomposition a single latent variable (LV) was estimated that—in a least‐squares sense—optimally represents the associations between age and subfield volumes. Thus, this LV represents a profile of subfield volumes with the strongest association with age. We assessed significance of the singular value corresponding to the LV via permutation test with 10,000 permutations. Robustness of within‐LV of subfield weights, that is individual contribution of each subfield volume to the LV, was assessed across 10,000 bootstrapped resamples of the data. By dividing each subfield volume's weight (from V) by its bootstrapped standard error, we obtained bootstrap ratios (BSRs) as normalized estimates of robustness (Figure 3A). BSRs are comparable to conventional z values, with values larger/smaller than ± 1.96 considered significant. This analysis also allowed us to calculate a summary measure of each participant's expression of the estimated LV by multiplying the model‐based vector of subfield weights (V) by each participants' vector of subfield volume estimates, producing a within‐subject value, which we termed a latent “age‐related HC profile” score. Finally, we assessed the association between the latent age‐related HC profile and sleep measures in 4 separate Pearson's correlations, with significance level corrected for multiple comparisons, with α = 0.05/4 = 0.0125 (Figure 3B). Entorhinal cortical volume was positively associated with objective sleep efficiency. Age groups shown were arbitrarily defined for illustrative purposes. Fourth, we used a similar PLSC to assess how the association between subfield volumes and sleep affects memory. For this analysis, we age‐residualized variables entered in the PLSC, because without this step any LV identified—expressing maximum common variance between two sets of variables associated with age—would be confounded by age. Here, on the subsample of participants (n = 55) having both hippocampal subfield and sleep EEG data, we first extracted four LVs that represent the maximum shared variance between a 4‐column matrix of subfield volume measures and a 4‐column matrix of sleep measures. Thus, these LVs represent profiles of subfield volumes (Figure 4A) with the strongest association with distinct profiles of the four sleep measures (Figure 4B). Again, we calculated a summary measure of each participant's expression of significant LVs by multiplying the model‐based vector of subfield weights (V) by each participants' vector of subfield volume estimates, producing a within‐subject value, which represents how strongly that participant expresses the pattern of subfield volume profile—sleep profile association captured by the LV. We termed this value an HC–sleep profile. Finally, we assessed the association between the LV and memory measures in 2 separate Pearson's correlations, with significance level corrected for multiple comparisons, with α = 0.05/2 = 0.025 (Figure 4C). (A) Subfields specific weights of the latent variable (LV) expressing the largest common variance between age and volumes of the four subfields. Bootstrapped values can be interpreted as z‐values, with values > 2 or < −2 (represented by the dashed line) considered significant. EC: Entorhinal cortex, Sub: Subiculum. (B) A latent age‐related hippocampal (HC) profile score (arbitrary units) with arbitrarily defined age‐groups. For each participant, this score reflects how much that individual's data represent the LV, that is the multivariate subfield volume—age association. This score is negatively associated with objective sleep efficiency. Finally, to investigate the effect of sex on our variables of interest and their interactions, we performed the following explorative analyses. We conducted t‐tests to detect differences between males and females in all variables of interest and conducted regressions to assess whether age associations interacted with sex, with significance level corrected for multiple comparisons, with α = 0.05/10 = 0.005 for both group of analyses. We performed regressions to assess whether sex interacted with any sleep measure to predict any subfield measure (4 × 4 = 16 tests, α = 0.05/16 = 0.0031), or to predict the age‐related HC profile score (4 tests, α = 0.05/4 = 0.0125). Finally, we performed 2 regressions to assess whether sex interacted with the HC‐sleep profile scores to predict memory (2 tests, α = 0.05/2 = 0.025). Results Age Is Negatively Associated With Hippocampal Subfield and Entorhinal Cortex Volume, Objective Sleep Efficiency, and Memory Bivariate zero‐order Pearson's correlations revealed significant negative age–volume associations (Figure 2B) for SUB (r (110) = −0.32, p < 0.001), CA1‐2 (r (110) = −0.27, p = 0.004), and EC (r (110) = −0.34, p < 0.001), but not DG‐CA3 (r (110) = −0.18, p = 0.055); a significant negative age–sleep associations for objective sleep efficiency (r (53) = −0.39, p = 0.002), but not for subjective measures of sleep quality or duration (all rs < 0.17, all ps > 0.139), as well as a significant negative age‐memory association for delayed recall on the RAVLT (r (115) = −0.31, p < 0.001). A full correlation table of our variables of interest is reported in the (Table S1). Objective Sleep Efficiency Predicts Entorhinal Cortex Volume Only a single model—objective sleep efficiency predicting entorhinal cortex volume—provided a significant fit to the data (F(9,37) = 4.54, p < 0.001, R 2 = 0.41) with objective sleep efficiency (β = 7.18, p = 0.004) and age (β = −4.04, p = 0.008) significantly predicting entorhinal cortex volume (Figure 3). Models for all other subfield volumes and sleep measures were nonsignificant (all Fs < 3.01, and all ps > 0.006). Model fit parameters are shown for all 16 models tested in Table S2, and statistics of the one significant model are shown in Table S3. Objective Sleep Efficiency Is a Specific Predictor of Age‐Related Integrity of Hippocampal Subfields and Entorhinal Cortex Our PLSC analysis identified a significant latent variable (LV), p < 0.001, that expresses the largest common variance between age and volumetric measures. Bootstrapped resampling of the data revealed that all volumetric measures contributed significantly (all ps < 0.029) to the LV with entorhinal cortex and subiculum (for both z > 3.9, p < 0.001) driving the LV–age association most strongly (Figure 4A). Pearson's correlations revealed that individual HC‐profiles were significantly negatively associated (r (53) = −0.42, p = 0.001) with objective sleep efficiency (Figure 4B), but showed no associations with any of the subjective sleep measures (all rs < 0.16, all ps > 0.25). Positive Sleep—Subfield Volume Coupling Is Associated With Less Forgetting A single significant (p = 0.0174) LV was identified which expressed a positive association between sleep measures and subfield volumes, with entorhinal cortex (p < 0.001) and subiculum (p = 0.042) contributing significantly (Figure 5A) to a positive association with sleep measures, and LV‐correlations being significant for objective sleep efficiency and AIS (Figure 5B). Pearson's correlations revealed that individual HC–Sleep profile scores were significantly negatively correlated with forgetting, r (51) = −0.31, p = 0.023 (Figure 5C). For interpreting this finding, note that HC‐Sleep profile scores are higher for participants with larger subfield volumes, and highest for those participants who have larger EC and SUB volume, as those volume measures were identified as having the largest weight (Figure 4A). Also, participants with higher score express the pattern of HC subfield volumes that captures the highest common variance between sleep and subfield measures. Thus, the negative association between HC‐Sleep profile scores and forgetting means that participants whose higher sleep quality is linked to larger subfield volumes show less forgetting. (A) Subfield specific weights of the latent variable (LV) expressing the largest common variance between age‐residualized volumetric and sleep measures. Bootstrapped values can be interpreted as z‐values, with values > 2 or < −2 (represented by the dashed line) considered significant. EC: Entorhinal cortex, Sub: subiculum. (B) Correlations of sleep measures with the LV in the optimal solution of the PLSC. Red lines depict 95% confidence intervals. Note that lower scores on AIS and GSQS mean better sleep quality. (C) HC‐Sleep score (arbitrary units) with arbitrarily defined age groups. For each participant, this score reflects how much that individual's data represent the LV, that is the multivariate subfield—sleep association. This score is negatively associated with forgetting after a 30‐min delay. Exploring Sex Effects on Subfield and Entorhinal Cortex Volume, Sleep, Memory, on Their Associations, and Interaction With Age Female and male participants only differed in memory performance (t(115) = 2.74, p = 0.007 for delayed recall and t(115) = −2.31, p = 0.023 for forgetting, all other ts < 1.44, and all other ps > 0.15), but no differences survived multiple comparisons. Regression analyses showed that sex did not interact with age in predicting any subfield, sleep, or memory measure of interest (all ps > 0.484). Further, sex did not interact with any sleep measure to predict any subfield measures (all ps > 0.028), or to predict the age‐related HC profile score (all ps > 0.052). Finally, sex also did not interact with the HC‐Sleep profile scores to predict memory (ps > 0.092). Discussion In sum, multi‐night sleep EEG and high‐resolution structural MRI of the mediotemporal lobe revealed that objective sleep efficiency were associated entorhinal cortical volume in a sample of cognitively unimpaired older middle‐aged and older adults when controlling for age. Importantly, age‐related differences in subfield volumes—mainly driven by entorhinal cortical and subiculum volume—were associated with objective sleep efficiency, but not with self‐report measures of sleep. Moreover, at the individual level, participants characterized by a common multivariate pattern of subfield volumes that contributed to positive subfield volume—sleep associations, performed better on a standard delayed recall task. Our finding that entorhinal cortex (EC) volume is associated with sleep efficiency is consistent with at least two other studies using similar high‐resolution MRI to assess mediotemporal lobe subfield volumes and at‐home polysomnography to assess sleep quality (André et al. 2023; Gervais et al. 2023). Gervais and colleagues reported a similar positive association between sleep efficiency and EC volumes in middle‐aged women who had undergone ovarian removal potentially due to direct effects of menopausal hormonal changes (Gervais et al. 2022, 2023; Mosconi et al. 2021). Andrée and colleagues found a negative association between EC volumes and sleep apnea severity in amyloid‐positive individuals. Based on studies detecting accelerated neurodegeneration (Daulatzai 2015) and elevated amyloid deposition (André et al. 2020; Yun et al. 2017) in individuals with disturbed sleep, these authors proposed that repeated hypoxic events may accelerate the neurodegenerative effects on EC and hippocampal subfields (however, see Moon et al. 2024). Our results, together with these studies, corroborate the suggestion that the EC is highly vulnerable to neurodegeneration induced by compromised sleep. In the current study, we did not assess amyloid burden, tau deposition, or menopausal status of participants. Therefore, we cannot confirm or exclude the possibility that lower sleep efficiency in our sample exacerbated hormonal effects or neurodegenerative effects on the entorhinal cortex. However, given that the association of sleep efficiency with EC was observed in a full sample of community dwelling participants, with no significant sex interactions, we speculate that decreased sleep efficiency may accelerate age‐related volumetric loss in otherwise healthy individuals. This is further supported by our finding that age‐related differences in hippocampal subfield and entorhinal cortical volumes were associated with objective sleep efficiency. Importantly, our results suggest that subfield volume–sleep couplings manifest in memory in cognitively normal older middle‐aged and older adults: participants exhibiting a pattern of larger volumes associated with better sleep quality performed better on a standard verbal episodic memory task. Critically, if these effects are not fully explained by neurodegenerative disease progression, it remains an open question whether they can be reduced or even reversed. Thus, future studies need to test whether sleep interventions can mitigate the effects of age‐related mediotemporal lobe atrophy (Falck et al. 2020). An intriguing result of our study was that the association of sleep quality with EC, as well as with age‐related integrity of hippocampal subfields and the entorhinal cortex, was observed only for EEG‐derived objective measures, but not for self‐report subjective measures of sleep. Again, this is consistent with other studies using highly reliable measures of hippocampal subfields and sleep (André et al. 2023; Gervais et al. 2023), who also reported associations only with objective sleep measures. Differences in reliability may be key to the discrepancies between the associations observed with objective and two subjective measures in our study, the Athens Insomnia Scale (AIS) and subjective sleep efficiency. While these subjective measures were administered once, we collected an average of six data points for objective sleep efficiency. Although we did not formally test the reliability of our measures, the increased number of observations likely resulted in greater reliability, which in turn, enhances the probability of detecting an association with another construct (Hedge, Powell, and Sumner 2018). More importantly, however, subjective and objective measures of sleep have been shown to be loosely correlated in older adults, given their differences in reliance on noisy memory processes, executive functions, the temporal window covered, and bias (Landry et al.). Overall, our results corroborate earlier recommendations to include both objective and subjective sleep measures to assess multiple aspects of sleep. Together, these results provide converging evidence for the notion that sleep is associated with age‐related changes in brain–behavior couplings. It also showcases the importance of considering objective sleep parameters for studies of these associations. Critically, many of the reported sleep associations were detected via sleep EEG recorded over multiple nights but not with self‐report measures. Thus, our results also highlight the sensitivity of an easy‐to‐use tool with high ecological validity that could provide objective measures of sleep quality. Some limitations of our study are worth noting. First, due to concerns of validity (Wisse et al. 2017), and in line with our prior work across the lifespan (Bender et al. 2018; Keresztes et al. 2020), we only assessed hippocampal subfield volumes along the hippocampal body, and assessed entorhinal cortical volume on six slices anterior to it. Second, compared to gold standard polysomnography, DH recordings, albeit easy‐to‐use and require less resources, provide less reliable data, both in terms of accuracy and data loss due to technical difficulties. Due to the related validity concerns of sleep architecture data, it remains to be established by future studies what—if any—changes in sleep architecture drove the observed subfield–sleep efficiency associations. Third, although we screened participants for diagnosed sleep disorders, we did not confirm exclusion of participants with sleep apnea with a questionnaire, and the Dreem system used in our study does not provide measures of apnea‐hypopnea. Thus, it remains a possibility that the detected sleep associations were driven by participants with untreated sleep apnea. Fourth, the absence of menopause assessment, including the reporting of menopausal hormone therapy in our study prevented us to assess potential hormonal effects on hippocampal subfields and entorhinal cortex, sleep, or their observed associations. Finally, the cross‐sectional age‐associations presented here cannot allow inferences about change and sleep–brain couplings over time (Keresztes et al. 2022; Raz and Lindenberger 2011). Conclusion To conclude, the study presented showcases that objective sleep measures can reveal otherwise unnoticed associations that are potentially important contributors of age‐related differences in brain‐behavior couplings. Future studies are needed to reveal subfield specific mechanisms driving these associations. In addition, our results call for longitudinal studies to allow for inferences about actual change. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Data S1. Acknowledgments This research was funded by the Hungarian National Research, Development and Innovation Office (grant RRF‐2.3.1‐21‐2022‐00015, as well as individual grants to A.K. [FK128648], P.S. [FK 142945] and N.B. [TKP2021‐EGA]), by HUN‐REN (grant 0708‐21 515 AT), and the Post‐COVID Programme of the Hungarian Academy of Sciences (to A.D.). A.K. was supported by a Max Planck Partner Group from the Max Planck Society. A.K. and P.S. were supported by the Bolyai János Research Scholarship of the Hungarian Academy of Sciences. Funding: This research was funded by the Hungarian National Research, Development and Innovation Office (grant RRF‐2.3.1‐21‐2022‐00015, as well as individual grants to A.K. [FK128648], P.S. [FK 142945] and N.B. [TKP2021‐EGA]), by HUN‐REN (grant 0708‐21 515 AT), and the Post‐COVID Programme of the Hungarian Academy of Sciences (to A.D.). A.K. was supported by a Max Planck Partner Group from the Max Planck Society. A.K. and P.S. were supported by the Bolyai János Research Scholarship of the Hungarian Academy of Sciences. Attila Keresztes and Éva M. Bankó contributed equally to this work. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. References Associated Data Supplementary Materials Data S1. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.